class

You can use more compute in backtesting in ways that don't incease the chance of overfitting. If I test a trading strategy with ranges for several parameters, I like to also test the strategy with those ranges on random price data, with returns either resampled with replacement or from an IID multivariate normal distribution matching the original returns. If the Sharpe of the best parameter set on historical data is not much higher than the best Sharpe found on random data, that is a warning sign.

PROBABILITY ARBITRAGE: HOW TO BEAT POLYMARKET USING DERIBIT OPTIONS Trading "Bitcoin Up or Down" on feelings is a casino. Trading them through options math is a systematic business. The strategy is simple: Deribit knows the future better than retail on Polymarket. The options market contains the volatility models of market makers like Galaxy and Wintermute. Our task is to export this knowledge into the inefficient Polymarket order book. 1) The Fundamental Idea Polymarket Up/Down markets are essentially binary options > If Price > Strike: Pay $1 > If Price < Strike: Pay $0 The price (e.g., 55 cents) is the implied probability (55%) Polymarket is driven by the crowd. Deribit is driven by giants using complex volatility models. If the Deribit model shows a 60% probability of an upside move, but Polymarket trades at 50 cents, you have found a Positive EV trade with ~20% ROI. 2) The Math To find the fair probability, we use a modified Black-Scholes formula for binary options. We need the Probability of expiring ITM. Formula for P(Up): Variables: > F (Forward Price): Futures price > K (Strike): The target price on Polymarket. > T (Time): Time to expiration in years. > σ : The hardest part - Implied Volatility (IV) 3) The Data Pipeline You cannot just scrape IV from the Deribit interface because there are no options expiring in 15 minutes. You need to build a Volatility Surface. Algorithm: • Snapshot: Capture the entire Deribit options book every 5-10 seconds. • Fitting: Build a Volatility Smile curve using an SVI model or cubic splines. • Interpolation: Interpolate σ for our specific time and strike • Calculation: Plug the resulting σ into the (d2) formula to get the Fair Price 4) Execution and Risks Example Trade: • Model: Calculates N(d2) = 0.62$ • Market: YES shares trade at $0.54$. • Edge: 0.08 • Action: Limit buy Pitfalls: > Spread & Fees: Your model must account for friction. If Edge < 2-3%, the trade is unprofitable. > Drift: On 15-minute frames, Forward is close to Spot, but during high volatility, the difference is critical. Always use perpetual contract data to calibrate. > Latency: The bot must react within milliseconds of a Deribit book update. You are not guessing where Bitcoin will go. You are arbitraging the inefficiency between a trillion-dollar professional options market and a retail prediction market This is pure quant trading